Ageing

{{hallMeta}}

{{pxDataLead}}

{{pxDataFoot}}

{{pxProvLead}}

{{pxProvFoot}}

{{methodName}} — {{methodDesc}}

The four methods are alternative analyses of the same data, not settings: rankings and the aging-up tail differ between them. Method 4 is the default and the one used in the publication.

Each cell shows how closely two epigenetic clocks agree across donors; an epigenetic clock is a statistical model that predicts age from DNA methylation, in model units, not years. Red means two clocks rise and fall together, blue that one rises as the other falls, and related clocks are clustered side by side. This shows which clocks measure similar things and which capture different aspects of ageing.

Pearson r within tissue, unweighted mean across 9 tissues ({{clockTech}}); ordered by hierarchical clustering on 1 − r, and unaffected by the gene, method and tissue selectors. Strip colours mark the clock class.

Each dot is one {{featureNounLower}} with a value in both tissues, for the method applied to the page: x is its {{cs}} with age in tissue A, y in tissue B. Both axes share one scale; the dashed lines mark zero and the solid diagonal marks equal coefficients. Dots above the diagonal are higher in B, below it higher in A.

Features without a value in either tissue are counted and left out — never drawn at zero. Now: {{tvCov}}. The plot is wider than tall; both axes keep the same length per unit of ρ, so y = x keeps its 45° meaning, and each axis spans every included value. B − A is a descriptive difference between two coefficients; no test, significance or causal reading is attached.

Density view (default): the background is a two-dimensional Gaussian kernel density of every {{featureNounLower}} with a value in both tissues, each with the same weight; no background feature is drawn individually. It is computed on a 128 × 128 grid in coefficient units, so the smoothing is the same in the card and in Enlarge. Bandwidth, equal on both axes: the mean of the two robust standard deviations × n^−1/6 (Scott's rule in two dimensions) — {{tvHtxt}}. The four bands, from light to dark, are the parts of the plane holding the densest 99, 90, 70 and 40% of the features: each threshold is read from the density at the features themselves. They show where coefficient pairs concentrate — not confidence regions, clusters or significance thresholds — and their colours say nothing about the sign of the coefficients. Only highlighted features are drawn as points (selected, compared, set members, found, pinned); Points shows every feature. Search, the region selection and its list, Differences, export and Save to My sets always use every feature's own coordinates, so both views give the same features.

A summary column (Across tissues in the full dataset, Across tissues · clock-matched in clock-matched samples) is the stored per-feature summary, not a mean of the plotted tissues. Comparing a tissue with a summary that includes it is descriptive and not independent; no test of the difference is made. If A and B are the same column every point lies on y = x by construction, so no density is estimated.

Tissues A and B belong to this panel only and never change the page's Tissue. Drag on the plot or use Select region to pick a rectangle; Explore selection as a set and Save to My sets act only when you press them.

Network — a pathway overlap map, drawn in the visual language of EnrichmentMap but with no enrichment test behind it. Each node is a gene set of the collection shown; its area grows with the number of the set's members found in the data (radius ∝ √n, with a legible minimum). Its colour is exactly the Landscape value: the median {{cs}} of the members with a value in the page's tissue for the page's method, on the same diverging scale as the heatmap (grey: no member with a value — not zero). A link joins two sets when their canonical members overlap: Jaccard = |A ∩ B| / |A ∪ B| over deduplicated Ensembl gene IDs (ambiguous and unmatched identifiers excluded), with link width ∝ Jaccard. Links depend only on membership, so changing the tissue recolours the map without moving it. The default cut-off (Jaccard ≥ 0.05 and ≥ 3 shared genes) is an exploratory display choice, not a significance threshold; sets without a link are still drawn — no link does not mean no association. Organic uses a deterministic force layout from a fixed seed. Circular orders names around an ellipse; Radial places highly connected sets nearer the centre. These are display arrangements, not biological clusters. Changing layout preserves selection and fits the view; Reset layout restores the chosen arrangement and clears selection. Nothing here is a NES, p-value, FDR or pathway activity.

One row per gene set, one column per tissue. Each cell is the median of the {{cs}} with age of the set's member genes that have a value in that tissue, for the method applied to the page. Each gene counts once per set. The Across tissues column is the median of each member's stored multi-tissue summary — not an average of the tissue medians.

The colour scale is shared by every cell and centred on zero; rows are not rescaled. A hatched cell (NA) has no member with a value — it is not zero. The tooltip gives the interquartile range between members (their spread, not a confidence interval), members with a value out of the set's total, and how many coefficients are positive, negative and zero.

This is a descriptive summary of members' age associations. It is not pathway activity, an enrichment score or NES, and no p-value or FDR is computed. Hallmark sets: MSigDB 2026.2.Hs, matched to this dataset by exact symbol; ambiguous symbols are left out and counted.

Chromosomes 1–22, X and Y are drawn around the circle with arcs proportional to their GRCh38 length. Gene positions come from the GENCODE release 39 gene records (the annotation GTEx v10 used); all 58,988 genes of this dataset are found there. Each gene is one marker at its locus midpoint — transcripts are never counted. Pseudoautosomal genes use their chrX record. The 37 mitochondrial genes are listed in the counts and the table but not drawn (16.6 kb would be invisible).

Inner ring(s): the aging track — every measured gene with a value, grouped in 2-Mb genomic bins; each bin shows the median {{cs}} of its genes, with the gene count in the tooltip. One colour scale, centred on zero, is shared by all tracks. Click a chromosome to see each gene's own value in a linear view. Outer rings: one ring per group (selection, compared genes, sets), with group colours that are never used for coefficients.

Regional age-association profile (optional): for each window along a chromosome, N is the number of genes with a value, zeros included. Positive = Σ max(ρ, 0) / N is drawn outward in red and negative = Σ max(−ρ, 0) / N inward in blue — the same denominator for both, and one amplitude scale for every chromosome and track. Windows are centred every step (5-Mb windows every 1 Mb by default; 1, 2 and 10 Mb available), clipped at chromosome ends and computed chromosome by chromosome; each gene counts once per window, by its locus midpoint. No bar means no gene in the window; pale bars mean fewer than 5 genes — a display convention, not a statistical threshold.

The profile mixes how many genes lean one way with how strongly. It is a descriptive local summary of expression–age associations — not methylation, a probability density, enrichment, significance or the activity of a region — and it is unrelated to the published Ageing-up and Ageing-down tails. It always uses every mapped gene with a value; showing or hiding groups never changes it. The Across tissues uses each gene's stored summary.

No arcs connect genes: sharing a set is not evidence of a biological link. The Across tissues is the pipeline's stored per-gene summary, shown as a summary track, not as a tissue. miRNAs are not placed: no verified locus coordinates are available here for mature miRNAs, and a mature miRNA can come from several loci.

{{prCounts}}

Each {{featureNounLower}}'s profile is its {{cs}} with age in each of the {{prNT}} tissues, for the method applied to the page. The Across tissues is left out, because it summarises those same tissues.

Profile similarity is the Spearman rank correlation between the reference's profile and a candidate's, over the tissues where both have a value (ties get average ranks). +1: the two rank the tissues in the same order; −1: in opposite order.

Candidates sharing fewer than 10 tissues with the reference, or with a constant profile, are excluded and counted. The reference itself is excluded. Ties are ordered by the number of shared tissues, then by name. The value describes how alike the age-association patterns are; it is not co-expression, an interaction or evidence of regulation, and no significance is attached.

{{dpConv}}

Each compared {{featureNounLower}} keeps one colour on the Correlation map and in the per-tissue chart.

On the map the coloured ring identifies the {{featureNounLower}}; the dot inside keeps this map's tail colour — red above μ + 3σ, blue below μ − 3σ, dark grey otherwise — so the ring never changes what the colour means.

Hover, focus or tap a name or a point for its coefficient and rank: rank 1 = highest {{cs}} among all {{entityPlural}} with a value in the applied tissue for this method; ties share the best rank. A {{featureNounLower}} without a value in the applied tissue is marked “no value” and never drawn at zero.